OTOmer Faruk Tuna

Omer Faruk Tuna

*This is an algorithmically generated author record based on publicly available information.

Omer Faruk Tuna has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    3PS-RAN: A Real-Time Framework for Securing the O-RAN RACH Against DDoS Attacks Toward NextG

    by Nurullah Aksu, Irsat Saridas, Utku Gülen, Ramin Fuladi, Ömer Faruk Tuna, Semiha Tedik Basaran

    Even with the introduction of additional security features, the progression toward 6G and Open Radio Access Network (O-RAN) architectures expands security risks, increasing both the scale and complexity of mobile networks. Consequently, robust defenses against threats are crucial to ensure the secure and reliable operation of these evolving systems. However, Denial-of-Service (DoS) attacks continue to exploit weaknesses in the RAN signaling protocols. Specifically, the Random Access Channel (RACH) procedure remains vulnerable to flooding attacks that target resource management and the finite capacity of the CU-UE-ID list, effectively blocking legitimate User Equipment (UE) access. In this paper, the possible vulnerabilities during the RACH procedure within an O-RAN environment are investigated. Based on discovered vulnerabilities, various RACH flooding attack scenarios and intelligent attacks are designed to perform DoS/Distributed DoS (DDoS) attacks on the system. To counteract such attacks, a novel Three-Phase Security (3PS) mechanism is designed to enhance protection during the RACH procedure in the O-RAN architecture. In this regard, both statistical and machine learning-based novel detection algorithms have been developed. Then, the detected attacks are mitigated by enhanced RACH protocol steps. The adverse effects of the aforementioned attacks and the security improvements introduced by the 3PS mechanism are evaluated in a real-time setup using the OpenAirInterface (OAI) 5G platform. Experimental results demonstrate that the proposed framework improves the average connection time for legitimate UEs by a factor of 9 to 25 under heavy attack conditions. Additionally, system profiling reveals that the proposed mechanism introduces a negligible operational CPU overhead of less than 0.2%, while maintaining a remarkably stable memory footprint. Furthermore, this implementation confirms the practical relevance of the designed attack scenarios and the solution’s compliance with 3GPP standards within an O-RAN context.


  • Open Access

    On the Deployment of AI-Driven Solutions in Edge-Enabled Beyond 5G Scenarios

    by Jordi Pérez-Romero, Ömer Faruk Tuna, Oluwatayo Kolawole, Oriol Sallent, Irene Vilà, Nikolaos Bartzoudis, Elli Kartsakli, Joan S. Pujol Roig, Maria A. Serrano, Utku Gülen

    The evolution beyond the Fifth Generation (5G) of mobile communications systems is expected to exploit different advanced radio access techniques, such as distributed multiple input multiple output (D-MIMO) or network slicing, together with an extensive support of Artificial Intelligence (AI) and Machine Learning (ML) solutions for different aspects of network optimization. These systems will also benefit from the evolution of edge computing technologies through the availability of computing and storage capabilities that will span from the radio access network nodes to the cloud, forming an edge-to-cloud compute continuum. The availability of these features will provide the beyond 5G (B5G) and 6G networks with an increased flexibility for deploying the different AI/ML-based optimization functionalities, trading-off aspects such as delay requirements, computational complexity or availability of data for feeding the AI/ML models. In this context, this paper takes as a reference the evolved edge computing architecture of the VERGE project and presents a pipeline with the required functionalities for the lifecycle management of AI/ML models in edge-enabled B5G systems. This pipeline is particularized for the deployment of two specific AI-driven solutions used for capacity sharing in network slicing and for D-MIMO power control.


  • Open Access

    Security of AI-Driven Beam Selection for Distributed MIMO in an Adversarial Setting

    by Ömer Faruk Tuna, Fehmi Emre Kadan

    In distributed multiple-input multiple-output (D-MIMO) networks, beam selection is necessary to predict the best beam and radio units (RUs) to serve the users in an optimum way. Finding the best RU and beam requires measuring the downlink channel for all possible RU/beam pairs, which becomes a resource-heavy operation, especially at the millimeter Wave band. To overcome this problem, artificial intelligence (AI) solutions are investigated which aim to infer the best RU/beam from sounding the channel for a subset of RUs and beams. While fairly accurate AI models can be obtained, these models have some intrinsic vulnerabilities to adversarial attacks where carefully designed perturbations are applied to the input of the AI model. In this study, we consider four different adversarial attack methods that craft perturbations using gradients of the AI cost function under two different beam reporting scenarios considering sequential and one-shot reporting of reference signal received power values for all RUs and demonstrate their effectiveness over traditional methods by extensive simulations, showing the necessity of smart defense techniques. To this aim, we propose an effective mitigation solution based on scrambling of RUs against these kinds of adversarial attack threats and verify the efficacy of our solution via detailed simulations. The proposed defense method provides up to 10 dB better signal strengths at the user side by selecting more accurate RU/beam pairs under adversarial attacks.


  • Open Access

    Practical Adversarial Attacks Against AI-Driven Power Allocation in a Distributed MIMO Network

    by Ömer Faruk Tuna, Fehmi Emre Kadan, Leyli Karaçay

    In distributed multiple-input multiple-output (D-MIMO) networks, power control is crucial to optimize the spectral efficiencies of users and max-min fairness (MMF) power control is a commonly used strategy as it satisfies uniform quality-of-service to all users. The optimal solution of MMF power control requires high complexity operations and hence deep neural network based artificial intelligence (AI) solutions are proposed to decrease the complexity. Although quite accurate models can be achieved by using AI, these models have some intrinsic vulnerabilities against adversarial attacks where carefully crafted perturbations are applied to the input of the AI model. In this work, we show that threats against the target AI model which might be originated from malicious users or radio units can substantially decrease the network performance by applying a successful adversarial sample, even in the most constrained circumstances. We also demonstrate that the risk associated with these kinds of adversarial attacks is higher than the conventional attack threats. Detailed simulations reveal the effectiveness of adversarial attacks and the necessity of smart defense techniques.

  • Open Access

    Unreasonable Effectiveness of Last Hidden Layer Activations for Adversarial Robustness

    by Omer Faruk Tuna, Ferhat Ozgur Catak, M. Taner Eskil

    In standard Deep Neural Network (DNN) based classifiers, the general convention is to omit the activation function in the last (output) layer and directly apply the softmax function on the logits to get the probability scores of each class. In this type of architectures, the loss value of the classifier against any output class is directly proportional to the difference between the final probability score and the label value of the associated class. Standard White-box adversarial evasion attacks, whether targeted or untargeted, mainly try to exploit the gradient of the model loss function to craft adversarial samples and fool the model. In this study, we show both mathematically and experimentally that using some widely known activation functions in the output layer of the model with high temperature values has the effect of zeroing out the gradients for both targeted and untargeted attack cases, preventing attackers from exploiting the model's loss function to craft adversarial samples. We've experimentally verified the efficacy of our approach on MNIST (Digit), CIFAR10 datasets. Detailed experiments confirmed that our approach substantially improves robustness against gradient-based targeted and untargeted attack threats. And, we showed that the increased non-linearity at the output layer has some additional benefits against some other attack methods like Deepfool attack.

  • Open Access

    Uncertainty as a Swiss army knife: new adversarial attack and defense ideas based on epistemic uncertainty

    by Omer Faruk Tuna, Ferhat Ozgur Catak, M. Taner Eskil

    Although state-of-the-art deep neural network models are known to be robust to random perturbations, it was verified that these architectures are indeed quite vulnerable to deliberately crafted perturbations, albeit being quasi-imperceptible. These vulnerabilities make it challenging to deploy deep neural network models in the areas where security is a critical concern. In recent years, many research studies have been conducted to develop new attack methods and come up with new defense techniques that enable more robust and reliable models. In this study, we use the quantified epistemic uncertainty obtained from the model’s final probability outputs, along with the model’s own loss function, to generate more effective adversarial samples. And we propose a novel defense approach against attacks like Deepfool which result in adversarial samples located near the model’s decision boundary. We have verified the effectiveness of our attack method on MNIST (Digit), MNIST (Fashion) and CIFAR-10 datasets. In our experiments, we showed that our proposed uncertainty-based reversal method achieved a worst case success rate of around 95% without compromising clean accuracy.


  • Exploiting epistemic uncertainty of the deep learning models to generate adversarial samples

    by Omer Faruk Tuna, Ferhat Ozgur Catak, M. Taner Eskil

    Deep neural network (DNN) architectures are considered to be robust to random perturbations. Nevertheless, it was shown that they could be severely vulnerable to slight but carefully crafted perturbations of the input, termed as adversarial samples. In recent years, numerous studies have been conducted in this new area called ``Adversarial Machine Learning” to devise new adversarial attacks and to defend against these attacks with more robust DNN architectures. However, most of the current research has concentrated on utilising model loss function to craft adversarial examples or to create robust models. This study explores the usage of quantified epistemic uncertainty obtained from Monte-Carlo Dropout Sampling for adversarial attack purposes by which we perturb the input to the shifted-domain regions where the model has not been trained on. We proposed new attack ideas by exploiting the difficulty of the target model to discriminate between samples drawn from original and shifted versions of the training data distribution by utilizing epistemic uncertainty of the model. Our results show that our proposed hybrid attack approach increases the attack success rates from 82.59% to 85.14%, 82.96% to 90.13% and 89.44% to 91.06% on MNIST Digit, MNIST Fashion and CIFAR-10 datasets, respectively.


  • Open Access

    Closeness and Uncertainty Aware Adversarial Examples Detection in Adversarial Machine Learning

    by Omer Faruk Tuna, Ferhat Ozgur Catak, M. Taner Eskil

    While state-of-the-art Deep Neural Network (DNN) models are considered to be robust to random perturbations, it was shown that these architectures are highly vulnerable to deliberately crafted perturbations, albeit being quasi-imperceptible. These vulnerabilities make it challenging to deploy DNN models in security-critical areas. In recent years, many research studies have been conducted to develop new attack methods and come up with new defense techniques that enable more robust and reliable models. In this work, we explore and assess the usage of different type of metrics for detecting adversarial samples. We first leverage the usage of moment-based predictive uncertainty estimates of a DNN classifier obtained using Monte-Carlo Dropout Sampling. And we also introduce a new method that operates in the subspace of deep features extracted by the model. We verified the effectiveness of our approach on a range of standard datasets like MNIST (Digit), MNIST (Fashion) and CIFAR-10. Our experiments show that these two different approaches complement each other, and the combined usage of all the proposed metrics yields up to 99 \% ROC-AUC scores regardless of the attack algorithm.